On this Page : Feedback or No Feedback
What Hi-Fi should do
If a fine instrument is played well its complex and delicate tonal quality produces a strong emotional response in the listener, the effect is pleasurable and it is an important part of the live music experience.
However the same intensity of feeling rarely occurs when reproduced music is played. Why does this aspect of the sound become lost and what parts of the hi-fi system are responsible for losing it?
One clue is the fact that old 78 rpm recordings acoustically recorded and replayed preserve this emotional response in listeners, despite the limited bandwidth and signal-to-noise. So sound fidelity in the conventional sense seems unconnected with the emotional stimulus. The action of the electronics (specifically the amplifiers) is the real culprit.
Since the arrival of low distortion feedback amplifiers, their true flaws
have been hidden by distortion measurements that declare them to be nearly
perfect and beyond reproach especially when compared with other links in the
audio chain -like the loudspeakers. Opinions are divided on this, whilst
designers of integrated circuit amplifiers claim near perfection others disagree and in
fact the listening experience tells a different story :
So music enthusiasts have looked for alternative solutions and as new technology has failed them they tended to fall back on designs from the past. This explains the revival of valve amplifiers and also transistor designs that were leading-edge technology 35 years ago. Depending on viewpoint the revival could be seen as either a revolution or a hiccup in amplifier development - but it has not been confined only to home Hi-Fi.
Valve amplifiers have returned to the guitar and P.A. scene because users like their sound characteristics. In the rush for the "sounds of the past" other amplifier design methods that have been thought to give better sound (class A single ended no-feedback) are once again being used with valve and transistor circuits.
Is this reincarnation of forgotten designs really necessary to reproduce music with the greatest fidelity or has something important been missed- something that prevents modern high performance amplifiers from working in musical terms?
Feedback - how it works
There is little doubt that negative feedback is involved in these issues. It was a technique invented by Harold Black in 1927 to improve the linearity of telephone amplifiers to enable them to perform better at high power. The negative feedback method takes part of the output of an amplifier with very high open loop gain and reduces the gain down to a set operating level by applying it via a defined resistance to the input. When applied to the input of opposite polarity the two signals (input and feedback) cancel except for any errors. The errors are actively cancelled by the feedback action so that the output becomes a very precise amplified copy of the input - in other words the distortion is reduced and the gain is defined.
If the open loop gain is very high then the distortion can be made to be so low
that it is considered by many engineers to be non-existent. Feedback action appears
to work reasonably well over the entire audio range from 20 Hz
to 20 KHz and beyond.
However it is a fact that as the frequency rises so that one cycle lasts a significant proportion of the feedback loop's transit time (typically 900 KHz in a modern compact design, effectively a radio frequency) the feedback performance becomes inaccurate, going out of phase so that instead of subtracting it adds when applied to the input signal, causing positive feedback and instability. A further complication is that at such high frequencies impedance mismatch causes reflections that add even more time-displaced high frequency energy into the system.
Designing for feedback
Engineers arrange for the feedback to be sufficiently in phase for the
amplifier to remain stable but, if maximum sound quality is required, more
precision is needed. Also engineers do not currently design to minimise the
impedance matching problem so the reflections are ignored. The audio amplifier
is always designed as an audio device, without reference to how it uses
radio frequencies (RF) to work because of the negative feedback..
When considering these issues the question most often asked is "if the frequencies causing the problem are so high, how can they affect the audio frequency range and change the sound"? In fact the two ranges, the audio frequency range and the feedback control range, are inseparable from the point of view of the amplifier. There is a tendency to think of feedback as an effect closely related to the signal frequency. In reality the "loop speed" of the feedback system is a fixed frequency, it is the highest frequency that can be corrected by feedback without making the original signal worse!
Problems at higher frequencies
As the signal frequency rises, in addition to phase shift, the feedback signal becomes distorted because the system is not designed for the RF range and the circuit design features that increase the open loop gain (constant current sources and current mirrors) are less effective, so the circuit becomes non-linear.
The non-linearity causes the feedback error correction signal to demodulate into the audio range and affect the sound quality. In other words the ultra-low distortion amplifier (assumed to be nearly perfect when measured at audio frequncies) is not low distortion at its control frequency.
In this way, it is possible for a feedback amplifier with very low measured distortion to suffer significant sound quality problems. The nature of the sound quality change is variable and complex but the neutrality of the amplifier is the main loss, feedback problems always leave the amplifer with a specific sonic signature. The sound can be too harsh or too soft or be masked or slow - it can exaggerate digital brightness.
Feedback or No Feedback - Implications for Sound Quality
It may be thought that it would be better to redesign the amplifier to eliminate the feedback altogether - this is a solution being advocated by the valve revivalists. When the real problem is appreciated this fix seems to throw out the baby with the bathwater!
The other alternative proposed is to eliminate global feedback and rely only on local feedback loops to create sufficient linearity. Attractive though this may seem it is not the answer. The loss of global feedback removes an easy way of controlling the overall Q of the system. The separate feedback loops must then be controlled and they must be set up so that they cannot intermodulate each other - very difficult to achieve in practice.
Instead, precise application of global feedback can be made to make the amplifier work as an aligned network, it is the best way of making the amplifier work really well.
A simple analogy illustrates some unappreciated aspects of the feedback problem.
Standing between two parallel mirrors you see your own reflection repeated away into the distance getting smaller and smaller. Move off to one side and the reflections become clearer because the view is less obstructed by you, but you see less images as they veer off to one side out of view. To get the maximum number of reflections you need to be aligned with both reflections i.e. exactly in between them.
An amplifier sees a similar effect within its feedback loop, but it goes one step further. Repeated images are generated inside the negative feedback loop as the amplifier tries repeatedly to correct errors that to succeed with the correction would require much greater bandwidth. The original error is aggravated by the mis-aligned loop response time--how long it takes to make an uncorrectable correction.
For the amplifier the images are all the same size but they go back in time not space. When everything is perfectly aligned the amplifier should see only one image, the original, the time delayed copies disappear.
When this happens the amplifier's performance becomes "greater than the sum of its parts", its audio resolution increases and the background from which it projects its sound stage becomes silent. Most importantly, the amplifier exhibits an extraordinary ability to reproduce fine and complex tonal colour because its own sonic signature becomes uniquely neutral.
This precision alignment cannot happen by chance because quite a number of adjustments must be very accurately set to make it work, the alignment can only be obtained by fully understanding the network.
DNM First Principles
© DNM Design 2013
(Legal & Credits)