In this month’s Roger’s Corner, we will discuss the use of negative feedback (NFB) in amplifier circuits which has been a point of contention in the audio world for years. Some audio designers eschew it, notably those who design single ended triode (SET) amplifiers. I recall Jack Elliano of Electra-Print providing me with an analogy that using NFB is like flooring the accelerator on your car while simultaneously having your foot on the brake. Other audio designers use NFB with excellent results, but there are many more using NFB as a “band aid” not knowing how to properly apply it in an amplifier circuit. Nelson Pass wrote that one should use a few dB of NFB or a very large amount. He believed a moderate amount of 10 dB to 20 dB was worse than using zero NFB and caused more issues in circuits than they remedied. Others including Ralph Karsten, Bruno Putzeys, and even Roger, as he will tell you later, felt the same way.
If like me you are old enough to recall the audio era of the 1970s buyers focused primarily on two specs, power, as many watts as one could afford, and total harmonic distortion (THD), the less the better to make those watts sound good, theoretically anyway. Manufacturers back then could easily use NFB to make an amplifiers THD look great on paper enabling them to sell more amplifiers. Unfortunately, these products were not always stable in the field as the NFB was applied while disregarding the point where the amplifier would go into oscillation. The typical uninformed audiophile back then readily accepted the manufacturer’s claims that the lower the THD the better the sound as if it were an edict. Sort of like in ancient Egyptian times when Rameses II would declare, “So let it be written so let it be done.”
So how does NFB affect the sound and how much NFB, if any at all, is enough? If an amplifier’s distortion is reduced by applying a moderate amount of NFB (between 10 dB and 20 dB), then 2nd order and 3rd order harmonics that the ear finds pleasing are masked by a cascade of higher order harmonics (7th, 9th, etc.) created by the NFB. This occurs because the amplifier’s nonlinearity switches between oscillating and non-oscillating states as the input signal crosses a critical threshold. The ear is very sensitive to higher order harmonics, perceiving them as bright or harsh. To avoid introducing higher order harmonic distortion when applying NFB the circuit design must have high Gain-Bandwidth Product (GBWP). Insufficient GBWP causes the open loop gain to drop off rapidly at higher frequencies. When the gain falls, the feedback loop cannot properly compensate for the amplifier's inherent nonlinearities and as a result the NFB produces higher order harmonic distortion.
The effect of NFB on phase shift also needs to be addressed, although phase shift in amplifiers manufactured today is less an issue than it was years ago. Amplifiers that exhibit phase shift inherently introduce a tiny time delay. This time delay is increased when using NFB so that when an amplifier runs out of open loop gain at higher frequencies there is a phase shift between the input signal and the feedback loop. If phase shift reaches 180 degrees, the internal components and feedback loop shift the phase of the signal so much the feedback becomes positive and is no longer working against the input signal but reinforcing it. At this point the NFB no longer corrects errors, and the signal delays become too long resulting in the amplifier oscillating and destroying itself or the speakers. So, as it is an inexact solution, audio designers cannot just arbitrarily apply NFB to remedy distortion. Yes, the specs may look good on paper, but at what cost? It would be better to use much less NFB, or none depending on the circuit, and live with a little more distortion as opposed to having an amplifier that is unstable.
I owned a pair of Atma-Sphere M-60 amplifiers that used 1 dB of NFB. Now 1 dB is miniscule, but it is what Ralph Karsten deemed appropriate for the circuit. One day Roger and I were bench testing the amplifier and he asked if I would like to put a feedback switch on the amplifier to switch between 1 dB and 6 dB NFB to which I agreed. The feedback resistor on the M-60 is 2 mega ohm (Mohm) and dropping it to 1 Mohm increases the NFB to 6 dB. It also cuts distortion in half and doubles the damping. The results were clearly audible, and more importantly the amplifier remained stable. Today, in the Atma-Sphere Class D amplifiers Ralph Karsten uses 35 dB of NFB. I had both the M-60 and Class D amplifiers in my system simultaneously for comparison, and I could not tell a difference. I called Ralph and politely told him the Class D amplifiers made my M-60 amplifiers obsolete and saw no reason to keep them. While Ralph acknowledged the relevance of my comment, he would neither agree nor disagree with it.
Bruno Putzeys and Ralph Karsten are two examples of audio designers that figured out a way to use much larger amounts of NFB without jeopardizing the stability of the amplifier. So why is the practice not more popular? Well, the short answer is that it is incredibly difficult to execute safely. However, thanks to their and others research it seems the solution to this problem turned out not to be abandoning NFB, but to drown the distortion in it. Once loop gain surpasses 20 dB to 30 dB, the feedback loop becomes strong enough to suppress all harmonics simultaneously, including the harsh, higher order ones. In addition, as you push the NFB higher, the distortion products drop well below the noise floor, leaving nothing but the original input signal. So, based on this research and subsequent advances of NFB implementation, it should alleviate audiophile concerns about NFB such that it does not ruin the sound or the amplifier.
However, I know that injecting any amount of NFB in an amplifier circuit will always be a bad thing to some audiophiles and I know it’s an exercise in futility to convince them otherwise. Personally, I am quite impressed at how the use of NFB has evolved over the past decades. I think one must tip their hat to the innovative audio designers who played a role in that, dating back to Julius Futterman. Yes, that Julius Futterman, the man who designed output transformer less (OTL) amplifiers, specifically the Harvard Electronics OTLs that used 60 dB of NFB. You see, Julius Futterman filed the patent to use massive NFB in audio amplifiers in 1953 and the patent was granted in 1956. The first commercial amplifiers to use this much NFB and maintain stability were the Harvard Electronics H3 and later the H3A (the “3” and “3A” designations represented the apartment numbers in New York city where Julius Futterman lived and built the amplifiers). Roger was a big fan of Julius Futterman so I thought you might appreciate that tidbit of history.
FEEDBACK on Feedback
By Roger A. Modjeski
Negative feedback is a technique where the output of a circuit is compared with its input, and a small difference often occurs. This difference becomes just a bit of correction that modifies the instantaneous voltage that is about to journey through the amplifier. This new voltage is shaped just right so that when it goes through the non-linearities of the amplifier it comes out just like the input signal, perfect. When all is done properly, there is no penalty to pay for this. The input signal comes in on one terminal (the base of a transistor or the grid of a tube), the feedback is often applied to the terminal beneath it (the emitter or the cathode), and the "corrected signal" appears at the collector or plate. The feedback provides the "correcting information" in perfect time. The amount of correction is generally just a few percent. It is exactly equal to the open-loop distortion of the amplifier before the feedback loop was closed.
Some look at feedback and think that since the output is being compared to the input, only differences or errors generate any action by the circuit with negative feedback. This would be incorrect as this makes it sound like the detector is idle waiting for an error to occur. It is not, in fact, it is passing the main signal all the time. In the example from the above paragraph, the input simply passes through the comparator unchanged if no change is needed. It really does leave it alone, I promise.
Read The Full Article