Atma-Sphere Class D Mono blocks
- Thread starter Chorus
- Start date
While I like to poke a little fun at Ralph I applaud his willingness and ability to think outside rhe box. That is why he makes cutting edge tube amps. He is working proof that tube advocates are not anti-science.
I do think that class D amps at present are a lateral move at the present. I have no doubt they will get better.
Meanwhile you can get some Atma-Sphere magic for $ 5.4k.
I do think that class D amps at present are a lateral move at the present. I have no doubt they will get better.
Meanwhile you can get some Atma-Sphere magic for $ 5.4k.
Hi Ralph. I'm an old timer, looking to set up a main system for retirement...I hate class D - granted, a recurring refrain - and would only look at it if someone like you said they had solved the main problems (lifeless highs, space being more of a void; decay dropping off rear stage in an unnatural way, not in a way that our ears/minds perceive as propagating sound through/within dimension, etc...). I'm a hard-wired tube guy, but, like you, am not fundamentally biased against any form of technology (its all rearranged matter to me...).
Q's: Will they run Sound Lab 645 Majestics, or different horses for courses? You mention an integrated...will it have a tubed line stage in front of the class D output? And, how about a tubed phono in the integrated as an option?
Thanks, Mark
They will (although the rear panel controls will have to be set to different settings, as with any solid state amp as opposed to tubes on that speaker. But the real issue is power; unless you have a very small room I don't think the class D amp makes enough since the Sound Lab is about 30 Ohms in the bass. This means that any solid state amp (including class D) will make about 1/4 of their 8 Ohm power. That is why a 150 watt tube amp can keep up with a 600 watt solid state amp on this speaker.
How big are your baffles ? Don’t want to take the thread OT, so have you posted about them on DIYA at all?I am using speakers of my own design, open baffle, no crossover, 15 inch full range drivers (25hz to 14khz), and Raven M1 supertweeters with a first order crossover at 16khz. All about 8 ohms and 98db efficient. I have made a custom speaker EQ (to boost the bass on the open baffle) that I implement in Roon and in my media server. Very transparent, fast, wide banded, dynamic, bass does go down to about 25hz in my room. So it really is a full range speaker. Very easy to hear differences between components, cables etc.
I've had the Boazu about three weeks. I love the Boazu's design philosophy. It does have a very pristine and transparent sound, definitely musical. But it also sounds a bit smaller, less dynamic, bass is clear, but doesn't have the power of my purifi. I will need at least another month working with it to get the best out of it before I make any real judgements though. My opinion may change!
I had the Orchard Audio Starkrimson GaN amp in for trial. It did lots of things very well, but I ultimately preferred the Purifi in my system. I felt that the qualities that fell short with it were really a matter of my taste rather than empirical. And I think that a real hi-end audio designer could probably get better performance from GaN, hence my excitement hearing about Ralf's new amp.
The best amps for them has been the Purifi, the First Watt J2 and the Shindo Sinhonia - 40 watt push pull F2a tube amp (I had a full Shindo rig from turntable to amps for a time - miss that!).
We have a 5th order feedback loop of our own design. It is self-oscillating, similar in spirit to Bruno Putzeys' designs. Unlike his stuff of course we're using GaNFETs. They are so fast that the deadtime used is no longer about the on and off switching times of the transistors- they can easily switch at 60MHz, at least from what I've seen. I can't speak for other designs but in our design the distortion sources are in the encoding scheme and the deadtime used in the output section (which is mostly used for the choke in the output filter) and tend to result in lower ordered harmonics rather than higher ordered. We attribute this to the amp's smooth character; of course its considerably lower distortion than any tube amp.
All class D amps have a switching frequency.
All amplifiers (tube, solid state or class D) have a phase margin, which is to say that when you apply feedback to the amplifier, there is phase shift in the amplifier circuit that causes the feedback to be positive at some frequency.
This means you have to be careful to not exceed the phase margin of the amp when using feedback, else the amp will oscillate.
In a class D amp, you can take advantage of that fact, by intentionally applying so much feedback that the amp oscillates as soon as its turned on. The oscillation is then used as the switching frequency. This allows you to run a lot more feedback than you could in a conventional design. It also allows for a relatively simple circuit, although sorting out the feedback network has to be done correctly (the 'oscillation criteria'), since its values define the switching frequency in addition to how much feedback is present. In our amp this value is about 35dB.
In a conventional amplifier you run into a problem with feedback, which has given feedback a bad rap in high end audio. To support feedback, you have to have a lot of gain and a lot of bandwidth, called Gain Bandwidth Product. If there isn't enough, you will see the distortion rise with frequency since the feedback is decreasing with frequency. In addition, feedback introduces distortion of its own thru non-linear bifurcation of the input signal, usually occurring at the point in the amplifier circuit where the feedback is applied. This introduces higher ordered harmonics into the amplifier, and this fact has been known for well over 70 years; Norman Crowhurst was writing about this problem back in the 1950s.
But if you can add enough feedback and it is the same value at all frequencies (IOW, there is enough GBP to go around) you can clean up the mess otherwise generated by feedback.
The result is low distortion and no rise in distortion with frequency, and the same distortion spectra at all frequencies. This allows such an amplifier to be smooth in its presentation, since the higher ordered harmonics that cause harshness and brightness are absent or masked.
So that is why self oscillation is important.
I got it on wikipedia. But your explanation is better. Thanks Ralph.
G'day, Ralph.
Hope you are well.
Will the Class D amps run off universal power (i.e. can I use them in Sydney @230v)??
Dear Ralph,
I am confused. A Class D Atma-Sphere amplifier?
It is not April 1.
I am confused. A Class D Atma-Sphere amplifier?
It is not April 1.
What you are really saying is that enough feedback suppresses the distortion products it produces and further chops them into higher harmonics. The result is as Crowhurst described as a signal correlated “noise floor “ that is really an infinite number of distortion products. The process that creates the high order distortion cannot also clean it up...it just grinds it down into a kind of distortion “dust” that is ever present and moving with the signal and is therefore pernicious.
We're using a toroidal power transformer. It is dual primary and rated to 50Hz.
Its not April yet and yes, we've been working on this project a long time, over 5 years.It seems you get around this problem when you have enough feedback. This sort of thing was not possible in Crowhurst's day.
The devices needed to get the Gain Bandwidth Product along with a lack of excessive poles in the amplifier design didn't exist, and wouldn't for another 4-5 decades (Crowhurst was writing about this in the 1950s). Crowhurst correctly identified the problem: non-linearities at the feedback node. If you think about it, this is going to be either a tube or transistor. The feedback has to be processed by either one. For example, in a traditional tube amp the feedback is put to the cathode of the first stage of gain. In many solid state amps, a differential amplifier is used at the input, one base being for the input signal and the other input base for the feedback. This means that whatever non-linearities exist in those devices is incorporated into the feedback; quite simply distorting the feedback itself, so it can never be right.
Without the Gain Bandwidth Product, whatever feedback applied to the amplifier will be less as frequency is increased. If you graph frequency vs distortion, you'll see this behavior- the distortion is increasing with frequency! This results in exactly what you stated- its pernicious.
Class D offers a way around that issue since the feedback can be applied without an active device processing it, and you can have the Gain Bandwidth Product such that distortion vs frequency is a flat line. The only other way to do this is to have no feedback at all, but you obviously get a lot more distortion that way. Distortion obscures detail, it obscures your ability to hear into the rear of the sound stage. Its important to get rid of it, but of course there's always going to be some. For that reason, what distortion there is has to be innocuous to the human ear; the distortion spectra is important for this reason.
We're using a toroidal power transformer. It is dual primary and rated to 50Hz.
It seems you get around this problem when you have enough feedback. This sort of thing was not possible in Crowhurst's day.
The devices needed to get the Gain Bandwidth Product along with a lack of excessive poles in the amplifier design didn't exist, and wouldn't for another 4-5 decades (Crowhurst was writing about this in the 1950s). Crowhurst correctly identified the problem: non-linearities at the feedback node. If you think about it, this is going to be either a tube or transistor. The feedback has to be processed by either one. For example, in a traditional tube amp the feedback is put to the cathode of the first stage of gain. In many solid state amps, a differential amplifier is used at the input, one base being for the input signal and the other input base for the feedback. This means that whatever non-linearities exist in those devices is incorporated into the feedback; quite simply distorting the feedback itself, so it can never be right.
Without the Gain Bandwidth Product, whatever feedback applied to the amplifier will be less as frequency is increased. If you graph frequency vs distortion, you'll see this behavior- the distortion is increasing with frequency! This results in exactly what you stated- its pernicious.
Class D offers a way around that issue since the feedback can be applied without an active device processing it, and you can have the Gain Bandwidth Product such that distortion vs frequency is a flat line. The only other way to do this is to have no feedback at all, but you obviously get a lot more distortion that way. Distortion obscures detail, it obscures your ability to hear into the rear of the sound stage. Its important to get rid of it, but of course there's always going to be some. For that reason, what distortion there is has to be innocuous to the human ear; the distortion spectra is important for this reason.
Its seems you get around the problem?? That is the best answer you can give for an apparently (according to you) self-cleaning process? As there is no sentience in feedback to go back and clean up after itself as perhaps a human might do, this is on the face of it impossible. Feedback does what it does ad infinitum. Even though Crowhurst was writing about this a long time ago, he did a thought experiment about what would happen if you could keep piling on the feedback. He came to the conclusion that you would get what looks like noise, because it is essentially an infinite number of distortion products, but it would be correlated with the signal and therefore would not behave like true noise. True noise you can actually hear signal below (like tape hiss) but a correlated "noise" floor you cannot and this has therefore negative sonic consequences.
If you agree that feedback is taking not only signal but the original distortion from the circuit back around the loop then it is clear that you keep slicing and dicing the distortion into evermore and smaller pieces but it does not get rid of them. And you keep original signal going around the loop to make new initial harmonics as well. Are they then to low in level to hear? Hard to say but something is clearly audible because you can take two different ultra low distortion products and they STILL don't sound the same with the same recording.
How small are the distortions from jitter? How about cables? How small is too small to hear. Clearly this process of crunching the level of distortion down while exponentially multiplying the number of products is testing this hypothesis but let's not kid ourselves that anything is cleaned up. It is an attempt to sweep it under the rug and I would argue probably not that successful. None of that correlated "noise" floor is masked, how could it be...with what? You have virtually eliminated anything that could mask it.
Class D amps don't have jitter; they are not digital. When distortion is kept at a low enough point, the music itself will mask it. It needs to be -100dB or so for that to occur. If its not that low, then lower ordered harmonics (2nd and 3rd) will need to be present to mask the higher orders to prevent brightness and harshness otherwise being present.
'Slicing' isn't a good word here as it does not describe what's going on when feedback is used. 'Cancellation' and 'bifurcation' are better. Cancellation is the part where feedback is able to suppress distortion. Bifurcation is where feedback adds some distortion of its own in the process due to non-linearities at the feedback node. IOW feedback does not 'slice' up distortion to get rid of it.
Crowhurst is correct in stating that the harmonics generated by feedback become part of the noise floor. IME if this is going on, the noise floor cannot be penetrated by the human ear as is possible when the noise floor is actually just hiss (due to how the ear's masking princple operates). Keeping that in mind, the noise floor of most class D amps is a good 10-15dB better than that of a tube amplifier and some are closer to -20dB. Do you see where we're going with this?
Ralph, did your reading comprehension ability drop or you had too much wine? I wasn't saying Class D amps have jitter. I was giving examples of other, seemingly minute, processes that are audible.
" When distortion is kept at a low enough point, the music itself will mask it. It needs to be -100dB or so for that to occur. If its not that low, then lower ordered harmonics (2nd and 3rd) will need to be present to mask the higher orders to prevent brightness and harshness otherwise being present."
I am shocked you actually believe this. If that were the case then there would be a number of "blameless" audio products out there. However, as you used to be aware, they all still sound very different from each other and probably from what the recording itself sounds like (as if we could really know this for sure). In that case why not just give up making amps because Benchmark, Halcro and a few others have beat you to that magic level...no need to go further..except have you heard them?
Benchmark Media Systems AHB2 power amplifier Measurements
Sidebar 3: Measurements
www.stereophile.com
Halcro dm58 monoblock power amplifier Measurements part 2
Fig.6 is a spectrum of the dm58's output at 200W into 8 ohms, taken with a very linear signal generator: the National Instruments NI4451 PCI card that Miller Audio Research uses as one of the platforms for its QC Suite measurement software. The second, third, and fourth harmonics are all between...
www.stereophile.com
Classé Delta Mono monoblock power amplifier Measurements
Sidebar 4: Measurements I tested the Classé Delta Mono with my Audio Precision SYS2722 system (see the January 2008 "As We See It"). I preconditioned the amplifier by operating it at one-third the specified power into 8 ohms for an hour. At the end of that time, the chassis was cool, at 76.9°F...
www.stereophile.com
https://www.stereophile.com/content/musical-fidelity-ams100-power-amplifier-measurements and it does it in all Class A!
Soulution 710 power amplifier Measurements
Sidebar 3: Measurements To measure the Soulution 710, I used Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see the January 2008 "As We See It" and www.ap.com). I ran the 710 at one-third power into 8 ohms for an hour, which maximally stresses a class-A/B...
www.stereophile.com
Even this one was nearly all below -100dB
"and we began with 20% global feedback. I found that more than 20% global feedback made the bass sound overly tight and constricted somewhat the musical flow of my system; less than 20% feedback pushed the sound beyond liquid, into a place that bordered on being soggy: Sorrentino's suggested 20% proved ideal. I ran the M1.1's using their balanced inputs."
Based on your criteria, the 100% would be preferable by the listener, no?
I think your new home at ASR is waiting...
Similar threads
