Why I Choose to Design with
Discrete Components, Not Op Amps

Bill Hutchins
Chief Designer
LKV Research, LLC

A designer of my acquaintance urged me recently “to also discuss why you design with discrete transistors rather than taking the easy way out with op amps.” It’s a great question. So I am writing this short paper explain my preference for designs using discrete devices (i.e., transistors, resistors and capacitors).

Op amps are electrical amplifier circuits with two inputs, very high gain and usually one output. They are typically complex circuits that require negative loop feedback for setting useful gain.[1] Properly employed, negative feedback can greatly improve the performance of an amplifying circuit. But, I believe that circuits using discrete components and operating without negative loop feedback can achieve equally good or better performance while avoiding pitfalls that afflict feedback-based amplifiers such as op amps. The first section below is devoted to the characteristics and behavior of op amps. The second describes how discrete, zero-feedback designs can achieve excellent performance, better in some ways than that of op amp circuits.

I - Op Amp Characteristics and Performance

To function properly, the op amp circuit must make high levels of “open loop” gain (the gain the circuit makes without any feedback being applied). A portion of this open loop gain is taken from the op amp’s output and returned as negative feedback to the input in order to set the working or “closed loop” gain of the circuit. This closed loop gain is what is used to amplify the musical or other input signal. The basic principle underlying the op amp is that, by making a very large amount of open loop gain and returning a great deal of negative feedback to the input, the closed loop gain and other important parameters of the amplifier will depend entirely on the feedback. That is, open loop gain and feedback are the only determinants of op amp operating characteristics. All other characteristics, such as open loop distortion are irrelevant. Well resp0ected op amp designer Barrie Gilbert put it this way:
The dominant mantra intoned over the application of op amps goes something like this: Never mind what's inside that cute little triangle; the function of the overall linear block … is determined (almost entirely) by the passive components that are added to provide feedback.[2]

These days op amps are almost always built in the form of integrated circuits on small silicon chips (integrated circuit or “IC” op amps).[3] Augmented with a very few resistors and capacitors, such IC op amp chips can be (and often are) used as low power gain stages in audio amplifiers and preamplifiers.[4] Used this way, op amps save the designer the work and headaches of designing his or her own gain circuitry. They also tend to lower parts count dramatically and thereby reduce cost.

A designer can select from among a very large number of IC op amps available today, many of which have been designed at least to some extent with audio applications in mind. But a designer has very little or no ability to modify or tweak the internal circuitry of these devices. They tend to come with features that I believe, based on listening to and designing a fair number of op amp based phono and line preamplifiers, detract from ultimate musical performance. Here are some examples of what I mean.

1 - Op Amps Employ Complex Circuits.

If I choose to design with op amps, I have to accept the complexity their designers typically build into them. To make the high levels of open loop gain necessary for proper operation,[5] op amps use multiple amplification stages created out of many transistors and diodes, along with a capacitor or two and some resistors. All of these components have to be fabricated within the IC chip. At a minimum, most op amps will have two voltage gain stages (the input stage and the gain or integrator stage) and one current gain stage (the output stage), along with a lot of ancillary circuitry.

If I use one op amp as a single gain stage in, for example, a phono preamp, I am actually sending the musical signal through a complicated circuit composed of multiple gain stages. Thus what first appeared to be a single gain stage sitting on the circuit board in my phono amp design, turns out to be several such stages when we look inside the little rectangular chip on which the IC op amp is built. Things get even more complex if, as is the case with a phono preamp, our design calls for more than one gain stage. We can easily wind up sending our musical signal through 10 or more gains stages if we use 2 or 3 op amp stages in series.

From the designer’s point of view, this problem has become worse over time. Older types of op amps often offer the designer a limited access to some of the internal nodes of the device such as voltage offset trim and voltage amplification gain. This access affords at least some ability to adjust the op amp’s internal circuits. More recent designs, however, often have little or no access to internal adjustments.

2 - Op Amps Severely Constrain Parts Choice.

Using IC op amps means having to use the parts (diodes, transistors, capacitors, resistors, etc.) that the op amp’s designer has chosen to have fabricated within the IC chip, not those that will produce the most musical sound. For phono stages, for example, my ears tell me that the best, most natural sound is obtained using specific very low noise, discrete jfets, either the now obsolete Toshiba 2SK170 or the new version of that device, the Linear Integrated Systems LSK170. I can’t get these in IC op amps. Some have jfet inputs, but those jfets are not the types I believe sound best. Moreover, most op amps use bipolar transistors, to whose sound I am not partial, in some or all stages in order to maximize gain.

Similarly, capacitors and resistors used in IC op amps are those that can be fabricated within the silicon chip, thus excluding the polypropylene film capacitors and metal film resistors I like to use for best sound in LKV designs. Also, with IC op amps, I am limited to whatever device matching (if any) has been built into the chip. This problem is ameliorated to some extent because the various components are all built onto the same chip. This method of manufacture has a tendency to reduce the component to component variation. But, I am still left to guess what degree of matching has been obtained and cannot, for example, be assured that jfets, resistors or capacitors are matched to the very tight tolerances we employ at LKV.

3 - Op Amps Require Negative Feedback.

This feedback is necessary to set the gain of the op amp at a useful level and to improve several aspects of its performance. Specifically, negative feedback:

· Reduces distortion.
· Lowers output impedance.
· Increases input impedance.
· Improves power supply rejection ratio (PSRR).
· Extends bandwidth

As useful as negative feedback is, it comes with important issues and difficulties. I discuss several of these below.

a. Feedback necessitates high open loop gain and substantial open loop distortion.

When an op amp is operating with feedback, the total open loop gain is allocated between closed loop gain and feedback. It is a zero sum game. The two resistor feedback network sets the closed loop gain (the operating gain of the op amp which will be used to amplify the music or other input signal). The remainder of the open loop gain appears at the input of the op amp as negative feedback. Thus, if the op amp has 90dB of open loop gain and the closed loop gain is set to 40dB, there will be 50dB of negative feedback. If the open loop gain were to drop to 70dB, the gain would stay the same but the feedback would decrease to 30dB. Thus, the op amp circuit must make a lot of open loop gain if there is to be enough such gain to support both the desired closed loop gain level and a large amount of feedback.[6]

The downside of thus making high levels of open loop gain is that the gain comes with high levels of open loop distortion, which can be quite complex. Very high gain stages are inherently nonlinear and thus prone to distort the signal. For example, Barrie Gilbert calculates the BJT[7] differential input stage of a typical voltage feedback op amp will produce a 1% level of 3rd harmonic distortion with a sinewave input of only 18 mV. He points out that such 3rd order distortion increases as the square of the input voltage.[8] Thus, if the input voltage increases from 18 mA to 100 mA, a level often seen or exceeded in audio preamplification, the open loop 3rd order harmonic distortion will rise to about 30%.

Several factors will cause this simple harmonic distortion to become relatively complex and therefore more difficult to calculate and measure. The use of multiple stages adds magnitude and complexity to the distortion, as do the Class B and AB output stages often seen in IC op amps.[9] Moreover, as the input voltage rises, the phase angle of the op amp’s output will also rise, a phenomenon inconsistent with a linear amplifier.[10] At low and moderate levels negative loop feedback, while decreasing the amount of 2nd and 3rd order harmonic distortion, actually increases the amount of the unmusical higher order distortions (4th order and up).[11]

When an IC op amp is used to amplify musical signals which contain multiple tones, intermodulation distortion is added to the mix. The more complex the music, the greater the magnitude and complexity of the intermodulation distortion become. The net result is not only unacceptably high levels of distortion, but also the potential for distortion peaks that are considerably higher than the “average” (rms) level of the complex distortion.[12] These peaks are both potentially damaging to the music and hard to predict and measure because they depend on the ever changing complexity of the musical signal being amplified.

Additional distortion can arise from the presence of ripple and other disturbances on the power supply rails.

This “soup” of complex distortions can result in some surprisingly high average (rms) percentage levels of open-loop, harmonic distortion, easily well into double digits.[13] With musical signals, levels of intermodulation distortion will be higher still. Of course, these levels tell us nothing about how op amps will perform in real circuits because op amps are not intended to be used without feedback. They do, however, alert us that that for the op amp circuit to perform linearly, we need a mechanism that will be effective in reducing distortion at all relevant frequencies.

Negative feedback will not only set the gain for the op amp circuit, but will also reduce this complex distortion. But, the reduction is proportional to the amount of feedback. To achieve low levels of distortion, one must use a lot of negative feedback.[14] It is quite possible to measure the average (rms) level of distortion and calculate the amount of feedback required to achieve a target lower level of distortion. For example, 50 dB of negative feedback will reduce a 10% average (rms) distortion level to about 0.03%, a perfectly acceptable level. 50 dB feedback also seems to be enough to avoid the problem that lower levels of feedback actually generate high order harmonic distortion products.[15] However, it is possible that greater feedback levels may be needed to push peak levels of complex distortion down to below audibility. It is hard to know just how much more feedback (60dB, 70dB?) is needed because the variability of complex music means we will likely not know how large the biggest distortion peaks are.

There are a number of difficulties associated with the use of feedback to reduce distortion in an op amp. We now examine some of them.

b. At high frequencies op amps may not have enough open loop gain to support adequate feedback.

As noted above, the total open loop gain of an op amp must be allocated between closed loop gain and feedback. A two resistor network is used to set the level of closed loop (i.e., operating) gain, which remains constant at all relevant frequencies.[16] At this fixed level of closed loop gain, the amount of feedback will be determined by the level of open loop gain. As open loop gain increases, feedback increases. Conversely, feedback decreases as the open loop gain drops.

A problem arises because, in virtually all op amps, open loop gain decreases as frequency increases. While typical op amps will have a 100 dB or more of gain at very low frequencies, that open loop gain starts to drop off at 100 or 200 Hz.[17] The decrease in open loop gain continues at a rate of 6dB per octave as frequency rises. As the frequency increases and open loop gain decreases, the amount of negative feedback in the circuit necessarily decreases also. Accordingly, in op amps distortion will rise with increasing frequency, other factors being equal.[18]

In the best low noise op amps currently available, open loop gain drops to about 80dB at 10 KHz, a frequency well within the audio band. Will that level of gain provide the 50dB of feedback that we calculated above would be needed to reduce 10% open loop distortion to acceptable levels? In some cases, the answer is no. For example, in a 2 stage phono preamp designed for use with very low output moving coil cartridges, we will need about 35 dB of gain in stage one and 50 dB in stage two. But with an open loop gain of 80dB and the 50dB of gain required in stage 2, we will have only 30dB of feedback available. That is probably not enough to effectively limit 10% harmonic distortion, and certainly not enough to cope with the higher levels of intermodulation distortion sure to be produced when the op amp is fed a complex musical signal.

c. Many op amps lack sufficient high frequency open loop gain necessary for good high frequency phase response.
Along with some other designers, I like to see a bandwidth extending at least to about 200KHz. That frequency is well about anything the human ear can hear. But, we get better phase response and more natural sound in the high end of the audio band if the linear, low distortion bandwidth of an amplification device extends out to about 200KHz. If we have a roll-off in amplitude or significant non-linearities (distortion) much below 200 KHz we experience phase anomalies in the high frequency portion of the audio band.

Many op amps are not effective at providing such a wide bandwidth. Again, the problem derives from the decrease in op amp open loop gain as frequency increases. By the time we get up to 200 KHz, for example, both of the low noise op amps discussed above have open loop gains well under 60dB. If as shown above, the gain of the first stage in our hypothetical phonostage is 35 dB, we will have only about 20dB of feedback at 200 KHz. This low level of feedback is likely to be insufficient to cancel the high levels of open loop distortion commonly encountered in op amps.

d. Negative Feedback Can Itself Create Distortion.

Both high and low levels of feedback can cause distortion in op amps. If the designer is able to maintain high levels of feedback at high audio frequencies, that feedback may create additional distortion. The feedback signal that is returned to the op amp’s input contains the open loop distortion that appears at the output. This distorted feedback signal is injected negatively into the input stage of the op amp where it combines with the musical signal being fed to the op amp. Normally, this negative feedback will reduce the distortion appearing at the op amp’s output. But when the distortion level is high and the input signal has significant high frequency elements, the input stage will see and must amplify a challenging mixture of music and distortion. This challenge is particularly great for op amps having input stages composed of a differential amplifier constructed using bipolar junction transistors (BJTs).[19] Such input stages are highly non-linear and are common in op amps designed for low noise. Walt Jung has shown that applying the distorted feedback signal to an op amp’s input stage will cause additional distortion of the output signal under the following conditions:

· Non-linear BJT input stage.
· High amplitude distortion components in the feedback returned to the input at frequencies above the point where the open loop gain starts to roll off.
· Substantial amounts of feedback at such frequencies.

In this situation, the complex signal created by the combination of the musical input signal and the high level negative feedback causes the input stage to behave non-linearly, thereby creating both phase shift and distortion at the output of the op amp.[20] In non-technical terms, as the level and frequency of the input signal increase, the op amp’s feedback loop simply cannot keep up and the musical signal is damaged.

Ian Hickman has identified a similar mechanism in which high levels of negative feedback in combination with high signal amplitude at frequencies in the upper part of the audio band can create transient intermodulation distortion. He posits a cymbal crash at the very top of the audio band (i.e., 18-20KHz) in combination with an orchestral crescendo. The several tones embodied in this combination will create considerable open loop intermodulation distortion. This distortion is fed back to the input of the op amp where it mixes with the incoming musical signal to create a very complex waveform. Hickman explains that this waveform is likely to overtax the second stage of the op amp, causing intermodulation distortion to appear at the output. [21]

Ironically, low levels of feedback can also exacerbate the distortion problem. When only about 20dB or less of feedback is used, 2nd and 3rd order harmonics are reduced, but of course the amount of reduction is less than would be the case with large amounts of feedback. The greater problem is that negative feedback at these low levels has been shown actually to produce increased amounts of un-musical 4th order and higher harmonic distortion.[22] This phenomenon can produce significant levels of high order harmonics when the level of open loop distortion in the op amp exceeds 1%.[23]

Finally, there is an additional distortion-related problem with feedback that is specific to phono stages with RIAA equalization. Such phono stages require roughly 40dB more gain at low frequencies (i.e., 20Hz) than at high frequencies (i.e., 20KHz). This demand for very high gain at low frequencies can mean that there is inadequate open loop gain available to reduce low frequency distortion to acceptable levels.[24]

e. Feedback circuits can cause harsh clipping.

In an amplifier circuit without negative feedback, an increase in the output level results in an increase in distortion which, in turn, causes a decrease in the circuit’s gain. For a 1% rise in distortion there is an approximately 3% loss in gain. As the voltage limit of the power rails is approached, this process results in a “gentle sort of limiting” or soft clipping.[25] In a circuit where gain is set with feedback, a much less musical form of limiting, harsh clipping, occurs. As the output voltage increases, the gain is held constant by the operation of the feedback. The signal increases linearly until it hits the voltage limit of the power rails, where it top is simply sliced off, resulting in a very sudden, harsh clipping that does far more damage to the music than does the gentler limiting of the non-feedback circuit.[26]

f. Crossover Distortion is not reduced by negative feedback.

In a Class B output stage of an amplifier or op amp, crossover distortion occurs in the transition between the positive to the negative output devices (i.e., tubes or transistors). This form of distortion is, in effect, a brief, precipitous drop in gain as the signal traverses the switching point between the output devices. This sudden drop momentarily reduces the open loop gain of the circuit so that there is very little or no gain support feedback. Hence, as there is no negative feedback available when the crossover distortion occurs, a feedback circuit simply cannot reduce the crossover distortion.[27]

g. Feedback can cause frequency response anomalies in phono preamps.

Tomlinson Holman presented a paper on phono preamp design to the Audio Engineering Society in 1975. In it he identified a number of problems presented by the interaction between the variable impedance of moving magnet the phono cartridges and a feedback loop containing the RIAA network. He found that a high frequency “roll-off” in the amplitude of the phono amp’s output often occurred in phono amps that depended on feedback to maintain a high, uniform level of input impedance. He found that the culprit was the open loop gain which dropped as frequency rose. As the open lop gain ddecreased, so too did the feedback level, rendering the feedback insufficient to maintain high impedance in an input stage using bipolar transistors. Conversely, he also found in a few situations in which an interaction between the feedback loop and the input amplifier caused a high frequency increase in input impedance with a consequent frequency response “roll-up” at the high end of the audio band.[28] He proposed a circuit that uses large amounts of feedback to raise input impedance uniformly and eliminate these problems.[29]

h. Summary.

From the forgoing discussion, I think we can distill the following:

· Op amps used to amplify musical signals, particularly complex music with multiple, loud tones and high frequencies, will exhibit large amounts of open loop harmonic and intermodulation distortion.
· Negative feedback can greatly reduce such open loop distortion, but its effectiveness is substantially lessened as signal frequency rises and as the op amp’s closed loop gain setting increases. This occurs because the open loop gain needed to supply the feedback is finite and decreases as frequency increases.
· In some circumstances negative feedback can actually cause distortion in op amps. This can be true for both too little and too much feedback.
· Negative feedback cannot reduce the level of crossover distortion in Class B output stages.
· Negative feedback can cause frequency response anomalies in RIAA phono preamps when used with moving magnet cartridges.
· Clipping is more abrupt and harsher in circuits with negative feedback than those without it.
· Determining the level of distortion in an op amp used for music can be difficult because that level is heavily influenced by the complexity of the signal being amplified.

In short, while negative feedback can be very useful, it also has the potential to create quite a few problems that will detract from the music we are trying to amplify.
Does all of this mean that op amps cannot be used in good sounding audio circuits? No. But it does mean that the designer who would use op amps for high end audio circuits has a field littered with pitfalls and issues to traverse. Op amps can no longer be viewed as simple, all purpose gain blocks in which high levels of open loop distortion can always be remedied by applying very high levels of negative feedback. Rather the designer must be aware of the kinds of problems described above and must search for ways to overcome them. Suggestions for how to do so can be found in several of the references I cite. But the task is not an easy one.

Indeed, in recognition of the difficulty of the task, some of the engineers most expert in op amp design and use are now suggesting a revised approach. Barrie Gilbert has spent a large part of his professional life designing op amps, including some of those most useful in audio amplification. He has observed that in the original “ ‘op am paradigm’ the idea was to make the open loop gain so very high that the function [of the op amp, i.e., its gain, distortion, bandwidth, etc.] is determined solely by the external components [principally the feedback resistors].” In light of the failings of this original paradigm, Gilbert recommends a “second design style typified by audio power amplifiers” in which “the objective is to achieve almost perfect linearity without feedback, and then use a small amount of feedback to squeeze out the last gram of linearity.”[30]
Gilbert thus recommends making the open loop circuit nearly distortion free, so that very little feedback is needed. I agree, but would take the point one step further. Would the best results not be obtained if it were possible to make a circuit so low in distortion and noise that no feedback at all is required and all the problems associated with feedback are obviated? In the final section of this paper I will describe how, at least for preamplifiers, this objective can be achieved with discrete, zero feedback circuits.

To Be Continued in Part 2

References

1. Walt Jung, Op Amp Applications Handbook, http://www.analog.com/library/analogDialogue/archives/39-05/op_amp_applications_handbook.html
2. Are Op Amps Really Linear? Barrie Gilbert, http://search.xfinity.com/?con=betac&cat=web&q=Are+Op+Amps+really+Linear%3F+ +Barrie+Gilbert
3. Audio Distortion and Feedback, Nelson Pass, https://passlabs.com/articles/audio-distortion-and-feedback.
4. Valve Amplifiers, Morgan Jones, 4th Ed. (2012).
5. Op-Amp Audio: Realizing High Performance: Bandwidth Limitations, Walt Jung, Electronic Design, http://electronicdesign.com/analog/op-amp-audio-realizing-high-performance-bandwidth-limitations.
6. Analog Electronics: Analog Circuitry Explained, Ian Hickman (1990).
7. New Factors in Phonograph Preamplifier Design, Tomlinson Holman, http://search.xfinity.com/?con=betac&cat=web&q=New%20Factors%20in%20Phonogra ph%20Preamplifier%20Design%2C%20Tomlinson%20Holman .

________________________________________
[1] “Negative loop feedback” refers to feedback in which a portion of the output of a gain stage or a whole amplifying device such as a preamp is taken from the output and inserted back into the input of that stage or device in such a way that it reduces the level of the output. The modifier “loop” distinguishes this sort of feedback from others, including “emitter degeneration” which refers to a particular, very common method of configuring a transistor used for amplification. In the remainder of this paper all references to “negative feedback” and “feedback” should be understood to mean negative loop feedback, unless the context clearly requires a different meaning.
[2] Are Op Amps really Linear? Barrie Gilbert, p. 1. Full citations and/or web addresses for each reference are given at the end of this paper.
[3] Op amps can be, and occasionally are, built out of discrete components, both solid state and tubes. However, in this paper the term “op amp” should be understood to mean an IC op amp, that is, one built as an integrated circuit on a silicon chip.
[4] This paper focuses on the performance of op amps and discrete components in preamplifiers (phonostages and line level units), the types of amplifying devices I have designed for LKV Research. The issues regarding use of feedback in power amplifiers are somewhat different and are not addressed here.
[5] “Open loop gain” is the gain the op amp makes before feedback is applied.
[6] Walt Jung, Op Amp Applications Handbook, p. 1.23, Fig. 1-13. Full citations and/or web addresses are given at the end of this paper for each reference cited.
[7] Bipolar Junction Transistor.
[8] Are Op Amps Really Linear? Barrie Gilbert, pp. 6, 8.
[9] Audio Distortion and Feedback, Nelson Pass (2008).
[10] Are Op Amps Really Linear? Barrie Gilbert, p. 8.
[11] Audio Distortion and Feedback, Nelson Pass (2008).
[12] Audio Distortion and Feedback, Nelson Pass (2008). For simplicity sake, I refer to the “average” level of distortion. But, of course, overall level of distortion is properly measured and reported on an RMS (root mean square) basis.
[13] Are Op Amps really Linear? Barrie Gilbert, 6, 8; Audio Distortion and Feedback, Nelson Pass (2008).
[14] But, there are circumstances as described below in which large amounts of negative feedback can actually cause distortion.
[15] Audio Distortion and Feedback, Nelson Pass (2008).
[16] The closed loop gain cannot be greater than the open loop gain. Therefore, closed loop gain will remain level only until it reaches a point where it equals the open loop gain. Thereafter, both decline together, and there is no feedback.
[17] Walt Jung, Op Amp Applications Handbook, p. 1.13, Fig., 1-7 ( ).
[18] Valve Amplifiers, Morgan Jones, 4th Ed., p. 54 (2012).
[19] BJT input stages are very common in op amps used for audio purposes because they are quieter than such stages executed with jfets.
[20] Op-Amp Audio: Realizing High Performance: Bandwidth Limitations, Walt Jung, Electronic Design (1998).
[21] Analog Electronics: Analog Circuitry Explained, Ian Hickman, p.91-92.
[22] Valve Amplifiers, Morgan Jones, 4th Ed., pp.219-220 (2012); Audio Distortion and Feedback, Nelson Pass (2008).
[23] Valve Amplifiers, Morgan Jones, 4th Ed., pp.219-220 (2012).
[24] New Factors in Phonograph Preamplifier Design, Tomlinson Holman, p. 5 (1975).
[25] Analog Electronics: Analog Circuitry Explained, Ian Hickman, p.91.
[26] Id., Analog Electronics, p.91.
[27] Valve Amplifiers, Morgan Jones, 4th Ed., p. 54 (2012).
[28] New Factors in Phonograph Preamplifier Design, Tomlinson Holman, p. 2-3 (1975).
[29] As explained below, I suggest a means of achieving the same result without feedback.
[30] Are Op Amps really Linear? Barrie Gilbert, 9 (emphasis in original); accord, Analog Electronics: Analog Circuitry Explained, Ian Hickman, p. 92.