Quantifying class A quality.
I'm posting this topic in the class D segment of diyaudio for reseons that will become apparent shortly.
The holy grail of audio amplification is the class A linear amplification because EVERYTHING is amplified and there is no crossover distortion or delay in the time base. It does not matter if it is Single Ended or Push-Pull. Linear class A is therefore our reference.
Class D is a switch-mode amplifier, similar to push-pull, it uses two MOSFET's arranged in a Cascode topology (half bridge) with a high- and low side. One for the positive voltage swing (high) and the other for the negative voltage swing (low) which produce a variable duty cycle pulse width modulated signal which in turn is recovered as an analog signal via (typically) 2nd order LC-filter which is used to drive your loudspeaker just as class A/AB. The main difference between class A/AB and D is that class D produce a binary/discrete copy of the analog signal vs class A/AB which is a direct amplification of the analog signal. Class D is very much an analog amplification method spite some claiming it as digital, which could stem from the description of the MOSFET function as being digital due to either being on or off aka 1 or 0.
The title of the thread is - Quantifying class A quality, so lets tackle that right away.
We are looking at two different approaches of how to amplify a sine-wave. One is class A which we can view as direct and the other is class D which we can view as discrete copy. Under an ideal situation, both will deliver exactly the same signal quality at the output.
Before I approach the core of the topic, allow me to quote Mytek:
We can see that Mytek is trying to challenge class A by stating that "well executed Class D circuits can challenge the best Class A and sound amazing" - Where is the evidence ?
Lets start with comparing Mytek Brooklyn amp against Accuphase A-75 which is a class A amplifier.
S/N ratio: 121 dB < good
THD: <0.01% < good
DF: 400 < good
S/N ratio: 122 dB < good
THD: 0.03-0.07 % < good
DF: 1000 < outstanding
- Bonus, lets also look at NAD M22 which is based on the Hypex nCore NC400 module in the case of a readers protest should emerge.
S/N ratio: >120 dB <good
THD: <0.003% < outstanding
DF: 800 <outstanding
The THD range is typically accepted as:
2% - Most people can detect this.
1% Some people can detect this but is a minority group.
0.1% Typically too low to be detected and is this level is therefore considered good.
0.01% This is the audiophile level, no one can detect this.
0.001% is well bellow what we can consider problems, you are doing well if your amp has this THD. Even lower than this has no real implication and is reserved for does who have the need to brag.
Did we solve it, can we say that Mytek (Pascal) and NAD (Hypex) are in the class A range of quality factors ? - well, no. Going back to class A being direct and class D being a discrete copy and also being a switch-mode amplifier, there is something els we need to look at. We need to look at the time base of things. I don't know what components Pascal is using but I do know what NAD / Hypex is using and I want to look at the MOSFET which is the switching component here.
I've been dealing with understanding how to quantify class A performance through class D, when we actually can say that class D is on par with class A and for this, I have turned to consideration of what real time is. A friend of mine is working as a programmer and is dealing with time based functions on a daily basis and he had this to say: "In hyper-physics, real time, as in happening right now, 1ns (nanosecond) is used as the reference."
Before we look at the Fairchild MOSFET we will look at a specific area of class D, the PWM side of things. We have the ideal PWM and we have the real PWM signal. The ideal PWM is a perfect square-wave with 90 degree angles. Going from an ideal PWM to a real PWM, we have to look at how the MOSFET behave and for that we have time based reactions called: Turn-on and Turn-off delay, as well as Rise and Fall time. Finally we have under and overshoot which is the cause for something called dead time. All of these are hardware based issues which class D has to deal with.
- The Turn-on and Turn-off is a time based reaction, from the demand to react (signal turns on) to the mosfet actually reacting and vice versa for when it turns off.
- Rise and Fall time is dealing with how much time passes from signal turn on to saturation or the linear region of the MOSFET.
This comes from the Fairchild FDP52N20 data sheet:
The combination of turn-on delay and rise time amounts to (typically) 228 ns or 0.228 microseconds. That is just for the MOSFET to fully turn on.
Then there is the matter of deadtime to prevent a short between the high and low side. When the MOSFETS turn on and off, they produce a transient spike. This is especially critical since both MOSFETS can't be on at the same time due to the dead short situation which will fry them in an instant. The deadtime is therefor inserted and this can be anything between 10-100 ns. You should read this as a period where no information can be captured and is the majority of the Jitter issue in class D and digital audio as well.
228 ns is far from the ideal PWM and the concept of real time in hyper-physics. So here is my question to the community:
What do you consider to be acceptable real-time in relationship to class A's continuous amplification ? Can we accept the 100's of ns time error considering the S/N ratio and THD figures we get via Hypex and Pascal.
Note that there are far better MOSFETS such as the Infineon BSZ097N04 and I can't speculate on the MOSFET Pascla use.
One small observation regarding those switching times - you'll note the datasheet specifies the conditions in which those times were measured, among which is a 25 ohm gate resistor and a voltage of 100v across the MOSFET.
Does the Hypex module in that NAD M22 use the same value, and/or the same supply voltage (+/-50v resulting in 100v net across each FET)?
The gate resistors, R13 and R33 has a value of 2.2Ω.
PSU Supply line looks to be +/- 14.5 V (29V net). There is also a +/-17.5V line (35V net)
And 1ns reaction time correlates with 300MHz bandwidth - audio ends at 20kHz.
voltwide, distortion detectable level varies widely. Last week I played around an ugly tpa3251 post-fb mode pre-clipping behavior, it produces very noticeable distortions at levels .1-.5% of THD@1kHz(at low frequencies even worse). The same time I need a few % to surely hear single 2nd or 3rd harmonic. Pre-clipping behavior a big deal and usually underrated, and never specified.
Poor class-D. Trying to mimic class-A while class-A is not the holy grail of anything.
Guy's. 1ns is no more than a reference point. I am sure it can be argued against and already has, which is fine. My question and the whole reason for starting this thread was to get more than my own "opinion" about when or for that matter how class D can be said to have class A sound quality.
If you don't like class D, then that is fine too, all you have to do is walk away... lol, no one is forcing anyone to like or buy a class D amp.
What makes class A superior? Lowest THD I suppose. But there is more about what makes an amp superior. Efficiency for instance, weigth, price just to name a few. These debates always run in circles...
Since my quantifying class A quality opens up a whole new jar of interpretations, maybe the answer is not a specific time frame but rather point towards using MOSFETS and other crucial components that allow for fastest turn-on / turn-off, rise amd fall time, slowest delays possible and the least amount of dead time in order to achieve the best PWM signal possible. There is no "close enough" or "this will do". If it can be improved and is practical, then that should be the goal. But for now, 1ns is my reference and no one els has to use that, but at least I can in a way quantify things.
1 GHz = 1 ns
25kHz = 0.04ms or 40 000 ns
35kHz = 0.028571ms or 28 571ns
44.1khz = 0.002267ms or 2267.573ns
192kHz = 0.0052083ms or 5 208.3ns
384kHz = 0.0026041ms or 2604.1 ns
1.5MHz = 0.000666ms or 666.66 ns
- EDIT -
And by the way, if we are talking about lowest THD, the NAD M22 win over Accuphase class A - NAD is an order of magnitude better.
|All times are GMT. The time now is 10:11 PM.|
Search Engine Optimisation provided by DragonByte SEO (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
Resources saved on this page: MySQL 17.65%
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
Copyright ©1999-2019 diyAudio