FWIW, I believe identical signals sound identical. Thus, if an amp has a low enough distortion level, and that means any difference between input and output other than gain, it will have no sound of its own. Unfortunately, we rarely, if ever, achieve that. Now, I took a decent sounding amp (though not ultra low distortion) that was stabilized with a cap across the feedback, removed the cap, and applied the usual Cdom cap on the second stage. The overall measurements were essential identical, but the sound went obviously dead. Within a few seconds of listening to it, I knew something was horribly wrong. Ditching the Cdom cap, and going back to global HF feedback with a cap across the feedback resistor, made it musical again. Remember, the gross distortion and response numbers were essentially unchanged. I think this is an extreme case, but it shows how critical the feedback area is, and IMO, how critical the design of a servo is going to be. OTOH, I also agree that it can be done well, but it's just not a fifteen minute design project.
I agree Conrad that using phase lag compensation can trash the fidelity. Although it doesn't have to. It's a rather complicated area.
My experience is that it is easier (requires less expertise) to get a good sound if the forward loop gain is kept low and phase lag compensation is minimized (eg: small or no Cdom). In this case the dc loop gain is likely to be too low to get a small enough output offset with just a cap., so roender would require a dc servo.
It is also my experience that a high forward gain approach with phase lag compensation will beat the socks off the other if implemented expertly. Then the dc gain is massive so a cap does the job.
I guess the choice is up to roender and how ambitious he is. In my experience it is easier to implement an inaudible dc servo than a high feedback circuit that sounds wonderful. This may be a restatement of Lazy Cat's point.
My experience is that it is easier (requires less expertise) to get a good sound if the forward loop gain is kept low and phase lag compensation is minimized (eg: small or no Cdom). In this case the dc loop gain is likely to be too low to get a small enough output offset with just a cap., so roender would require a dc servo.
It is also my experience that a high forward gain approach with phase lag compensation will beat the socks off the other if implemented expertly. Then the dc gain is massive so a cap does the job.
I guess the choice is up to roender and how ambitious he is. In my experience it is easier to implement an inaudible dc servo than a high feedback circuit that sounds wonderful. This may be a restatement of Lazy Cat's point.
My view is no servo and no caps. This will usually require a bipolar input design (FETs tend to drift a bit more) + some sort of initial offset adjustment, but once done, the offset drift over temperature is very low. If you use a fully balanced front end topology, the N/P pair base currents cancel (especially if you match the Hfe's) further improving the situation.
Conrad, regarding your Cdom experiment, can you remember how much HF loop gain you threw away by using it, compared to the feedback roll-off method? IMHO the loss of HF loop gain (in the 2kHz...20kHz region) is the reason for the bad sound when to much Cdom is used.
@all: everbody is invited to participate in the DC-servo thread over at chip amps.
- Klaus
@all: everbody is invited to participate in the DC-servo thread over at chip amps.
- Klaus
Lazy Cat said:Roender, I know the amp is no negative feedback design, what I meant is that you've better lower open-loop gain to reasonable 60dB, to have app. +30dB gain excess over closed-loop feedback.
It's well known, the transient and intermodulation distorsion are worse with higer open-loop gain, explanation is simple: transient has a great frequency spectrum (Feurier analysis) and separate gain stage can not follow the transient-pulse signal correctly because of high gain and poor frequency response. All spectrum signals (high harmonics components) greater than frequency response of certain gain stage are lost and the signal is therefore distorted.
The art of amp designing is to find optimum open-loop gain to determine good correlation between slew rate (high gain - amp is faster and more unstable) and transient signal distorsion (low gain - more high harmonics because of greater freq. response).
95dB of open-loop gain is not reasonable for an audio amplifier. In audiophile community 30dB of open-loop gain or no negative feedback design is more prefered, because there are more harmonics in output signal and the sound is better.
This amplifier, shown in one of my previous message, have a single amplification stage and no VAS.
SR is over 70V/uS and the dominated pole is at around 3kHz. It is fast enough to not have much IM or TIM distortion.
Kle wrote:
Let me rename this the "feedback loop gain" (FLG) to be clearer. I would call a value of FLG >15dB as "high" in the context of audio power amp design. By "high" I mean things can get tricky.
This is because 15dB is the approximate gain above which stability problems can arise in ideal 2nd order (2 pole) linear NFB systems. It's the gain at which the phase margin is 45deg. And audio power amps are normally fair approximations to 2nd order systems over the frequency range being controlled from the point of view of feedback stability. Non-linearity changes things for the worse.
In the 1980s I read a magazine interview with Dan D'Agostino of Krell where he was talking about the KSA50. He said that they basically made a flat OL gain amplifier as linear as possible (class A) and then wrapped 14dB of global NFB around it. He said that above this value the benefits of the feedback were eroded.
The most rewarding approach for many is the Krell approach. It requires a lot of effort to get the OL distortion low but then you slap 15dB of NFB around it and it gets even better and is well behaved. This is what I'd call the "OL optimization" approach. The feedback makes it better but isn't an integral part of the design.
The other approach is what I'd call the "control system" approach. Here the feedback is integral to the design. If you cut the feedback path the design simply won't work. It's like those military fighter jets whose wing profile has to be computer controlled for them to fly. In return you get excellent responsiveness. You make something whose OL performance is nothing like the result you want to end up with. The FLG varies with frequency and doesn't have that much meaning in itself...it is other factors such as stability margin that matter more.
Let me rename this the "feedback loop gain" (FLG) to be clearer. I would call a value of FLG >15dB as "high" in the context of audio power amp design. By "high" I mean things can get tricky.
This is because 15dB is the approximate gain above which stability problems can arise in ideal 2nd order (2 pole) linear NFB systems. It's the gain at which the phase margin is 45deg. And audio power amps are normally fair approximations to 2nd order systems over the frequency range being controlled from the point of view of feedback stability. Non-linearity changes things for the worse.
In the 1980s I read a magazine interview with Dan D'Agostino of Krell where he was talking about the KSA50. He said that they basically made a flat OL gain amplifier as linear as possible (class A) and then wrapped 14dB of global NFB around it. He said that above this value the benefits of the feedback were eroded.
The most rewarding approach for many is the Krell approach. It requires a lot of effort to get the OL distortion low but then you slap 15dB of NFB around it and it gets even better and is well behaved. This is what I'd call the "OL optimization" approach. The feedback makes it better but isn't an integral part of the design.
The other approach is what I'd call the "control system" approach. Here the feedback is integral to the design. If you cut the feedback path the design simply won't work. It's like those military fighter jets whose wing profile has to be computer controlled for them to fly. In return you get excellent responsiveness. You make something whose OL performance is nothing like the result you want to end up with. The FLG varies with frequency and doesn't have that much meaning in itself...it is other factors such as stability margin that matter more.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.