Chapter 30

Theory

Theory

The theory of the Otala amplifiers, and thus the basis for the Electrocompaniet amplifiers, has been detailed in many papers, mostly by Matti Otala, but also by several others. At Electrocompaniet these theories were extended, and a theoretical framework for our amplifiers was established. The following pages give you an insight into both the original theories by Dr. Matti Otala, and also the theories built up in the first years of Electrocompaniet. Some of the thoughts are solely mine, however — don’t assume that Electrocompaniet of today will vouch for all I say here :-)

This is the original outline for the Theory section. Most items below were only ever sketched as topic headings on the old site and never actually written up; they’re kept here as a record of the plan, marked (not yet written).

On TIM, DIM, nonlinearity and distortion

  • TIM — this was the starting point of Otala’s theories, and his own idea from the beginning. In the early ’70s many people had already noticed that the sound of the new transistor amplifiers was inferior to the sound of tube amplifiers. How could that be, when the transistor amps had less distortion than the tubes?
  • DIM — what it is, and how it relates to TIM. This was when everybody tried to find measurement methods. (not yet written)
  • The debates and the quarrels — but Otala was right! (not yet written)
  • “All distortions are equal, but some are more equal than others.” That was the feeling at Electrocompaniet at that time, and the beginning of an understanding of how these distortion mechanisms interacted, and how they affected the sound — and that the basis of all distortion is nonlinearity. The nonlinearity is the source, the distortions the symptoms. (not yet written)

A theoretical framework for building good-sounding amplifiers

  • A theory of single stages — a single stage is the basis for all amplification, and at Electrocompaniet a model for the single stage, and “our way”, evolved.
  • A theory of multiple stages — multiple stages put together do not behave as N times a single stage; they interact, and in many cases even counteract. (not yet written)
  • A theory of feedback — many people are either pro-feedback or fully against feedback. At Electrocompaniet we learned to live with feedback in all forms, local and loop, and learned how to get it as a friend and not an enemy. (not yet written)
  • Output stages — that current, what does it do? (not yet written)

How do we calculate the nonlinearity

  • Calculations of input-stage nonlinearity — this is simple, once you know how; a table summarises everything.
  • Calculations of the nonlinearity of multiple stages. This is more complicated, but by following a set of rules, and abiding by the general model we use, it is not that difficult. (not yet written)
  • Calculations of distortions — one thing is the nonlinearity, another is how much distortion of the different types your amp will measure. (not yet written)
  • Designing for lowest distortion. Note that there very seldom exists one optimum point — there are just too many variables, so you have to use some creativity and a feel for what’s going on. (not yet written)

Frequency response, rise times, slew rate

What’s the difference?

  • Open and closed loop frequency response
  • How the frequency plane and time plane interconnect (not yet written)
  • Time-response behaviour (not yet written)
  • Poles and zeroes (not yet written)
  • Compensation of an amplifier (not yet written)

Other factors

  • The damping factor — or output impedance, what is it? (not yet written)
  • Why there are no simple solutions (not yet written)
  • Power-supply interaction (not yet written)
  • RFI — what can it do (not yet written)
  • The importance of the components (not yet written)

Subsections of Theory

A Theory of Single Stages

The general idea

When you design something, you’re moving your ideas and abstract thoughts into the real world. Using electronic components, they will never act exactly equal to your ideas. So the best approach is to approximate their behavior, and make them work as close to your ideas as possible. Further, if you choose your models skillfully, you will get a behavior from the device which very closely mimics what you’re after.

Therefore, the sequence is: Model ⇒ Design ⇒ Measurement ⇒ Listening — and you circle this sequence until you’re happy :-)

The walkthrough below is rather detailed, to show the general outline of the procedure. The other derivations are less rigorous.

A model for a single stage

We model a single stage as a voltage-controlled current source. In our ideal world this means we want an active device working as a transconductance device. A transistor (and also a tube) is very close to this ideal. But one has to further improve the circuitry around the active device, in order to make it behave as closely as possible to this ideal.

What this means is that a perfect transistor, and thus a perfect stage, will have infinite input impedance, no reverse coupling from output to input, infinite output impedance, and a finite and constant (with respect to both the signal and the environment) transfer conductance.

Real transistors are not quite as good as this, but we can improve on the transistor in order to make it behave more like this ideal. Doing this will normally make the stage perform better in all respects, but keep in mind that all rules will turn back on you at a certain stage. There is no such thing as a free lunch.

If such a stage is voltage-driven, we will reduce the nonlinearity from all “leakages” back to the input, be it input impedance or reverse coupling. The dominating nonlinearity will then be the transconductance, which is easy to control.

A transistor in a common-emitter coupling is the starting point. It has, in principle, the behavior described above. The following rules exist:

Linear behavior:

Transfer conductance, given by:

$$g_m = \frac{I_e}{V_T}$$

$I_e$ is the emitter DC current and $V_T$ is the voltage equivalent of temperature, normally equal to 25 mV — the exact formula is $kT/q$, where $k$ is Boltzmann’s constant, $q$ is the charge of an electron, and $T$ is the absolute temperature in Kelvin.

The inverse of the transconductance is called the dynamic resistance, called $r_e$.

Current amplification: $H_{FE} = I_c/I_b$, derived for a particular current $h_{fe} = i_c/i_b$, which applies for small-signal currents around a quiescent point $I_c$.

Input impedance: $r_{in} = h_{fe} \cdot r_e$

The dominating nonlinear mechanism lies in the transconductance. Since this is a single stage (not a differential stage) it will generate a smooth series of harmonics (if stimulated with a pure sinusoid).

If the input signal is given as $x$ (where $x$ can be e.g. $\sin(\omega t)$), then the output $y$ will be:

$$y = a_1 x + a_2 x^2 + a_3 x^3 + \cdots$$

And $x$ is a relative parameter which must always obey $|x| < 1$ in order for the series to converge. If $|x| < 1$ then it follows that $|y| < 1$. The output current is $i_e$ and the output parameter is then $i_e/I_e$. The input signal generating a current of $i_e$ is $u_{in}$, from the formula $i_e = u_{in} \cdot g_m$. It then follows from the transconductance formula that the input parameter we seek is $u_{in}/V_T$.

So, given the input $x$, defined as $u_{in}/V_T$, and the output $y$, defined as $i_e/I_e$ — what do these things mean?

If $u_{in}/V_T$ exceeds 1, then the varying part of the current exceeds the quiescent current $I_e$, and the stage is clipping. When the stage is clipping, our formula breaks down. If we want to find the distortion when the stage is clipping, we’ll have to resort to Fourier analysis.

To make this into a practical case: assume a transistor running at a current of 1mA. The $g_m$ is then 1/25 siemens (the inverse of ohm, the unit for transconductance, although “mhos” — ohm reversed — is also used). An input signal of 1 mV will then generate an output current of 1/25 mA = 40 µA. But now note: this is the first-order approximation. As can be seen from the series expansion above, we also have second- and third-order components. The first-order coefficient should be pretty close to $g_m$, but what are the other two coefficients?

The real equation relating input voltage to output current is $I_e = I_s \cdot \exp\left(\frac{U_{be}}{V_T} - 1\right)$, called the Ebers-Moll equation. $I_s$ is the “leakage” current, but don’t bother about it — we’ll soon enough get rid of it. We are interested in finding the equation for the behavior around the quiescent point. We do this by adding small deviations $i_e$ and $u_{be}$ to the equation above. Resolving this, we get the much simpler equation: $i_e/I_e = \exp(u_{be}/V_T) - 1$, and its inverse: $u_{be}/V_T = \ln(1 + i_e/I_e)$. These equations are called the signal equations, and will be used to get the coefficients for the series expansion above. We’ll first make a series expansion of the first equation, and we get:

$$\frac{i_e}{I_E} = \frac{u_{be}}{V_T} + \frac{(u_{be}/V_T)^2}{2} + \frac{(u_{be}/V_T)^3}{6} + \cdots$$

The efficiency parameter

We now introduce the efficiency parameter $n_i$. The point of introducing this parameter is to generate simpler formulas for calculating the distortion, and to gain a better understanding of how the distortion and other transistor parameters are coupled.

The parameter is defined as:

$$n_i = \frac{i_e}{I_e}$$

For a bipolar stage without local feedback, as the stage discussed above, $n_i$ is equivalent to $u_{be}/V_T$, where we only consider the linear part of the series expansion. The equation above can then be written as:

$$\frac{i_e}{I_E} = n_i + \frac{n_i^2}{2} + \frac{n_i^3}{6} + \cdots$$

The second-order distortion is defined as the second-order term divided by the first-order term, and the third-order distortion in the same manner:

$$\text{2nd} = \frac{n_i}{2} \qquad \text{3rd} = \frac{n_i^2}{6}$$

If the input signal is a sinusoid, then the following equations hold (ask via the about page if you’d like the proof), where $\text{2ndh}$ is the second-order harmonic distortion, and $\text{3rdh}$ is the third-order harmonic distortion:

$$\text{2ndh} = \frac{\text{2nd}}{2} \qquad \text{3rdh} = \frac{\text{3rd}}{4}$$

Putting it together:

$$\text{2ndh} = \frac{n_i}{4} \qquad \text{3rdh} = \frac{n_i^2}{24}$$

…and remember, $n_i = u_{in}/V_T$.

FET stages

It is interesting to do the same exercise for FET transistors. The result is a simpler series, with only first- and second-order components. By inserting the efficiency parameter, one gets the following equation for the FET’s second-harmonic distortion:

$$\text{2ndh} = \frac{n_i}{8}$$

Half the amount of the bipolar transistor. Some people have argued that the FET is a much more linear device than the bipolar. This equation shows that to be only a partial truth — there is only a 6dB improvement.

Local feedback

It is well known that local series current feedback (read: inserting an emitter resistor) reduces the distortion. The feedback factor can be written as:

$$D = 1 + g_m R_e = 1 + \frac{I_e R_e}{V_T}$$

and the resulting efficiency/distortion equations are then:

$$\text{2nd} = \frac{n_i}{2D} \qquad \text{3rd} = \frac{n_i^2}{3D}$$

or for the harmonic distortion:

$$\text{2ndh} = \frac{n_i}{4D} \qquad \text{3rdh} = \frac{n_i^2}{12D}$$

Note that the efficiency parameter $n_i$ is still defined as $i_e/I_e$, but the input version is now $n_i = u_{in}/V_{th}$, where $V_{th} = V_T D$.

Differential stages

There is a similar set of equations for the differential pair.

If the stage is completely in balance (which of course rarely happens), all second-order components will be cancelled.

The no-feedback solution will have a basic transconductance of $g_m = I_e/(2V_T)$, which gives $n_i = u_{in}/2V_T$.

The corresponding distortion is then equal to the single stage, except for a mismatch parameter $m$ and a common-mode signal factor $c$:

$$\text{3rd} = \frac{n_i^2}{3} \qquad \text{2nd} = \frac{(m+c) \, n_i}{2}$$

If the current source feeding the emitters has infinite output impedance, $c$ approaches 0. The formula for $c$ is $c = i_k/(2i_e)$. More information on this is in my AES paper (not yet ported to this site). There used to be an ActiveX Single Stage Calculator here as well — it’s retired along with the rest of the site’s old ActiveX controls.

Summary

Stage type2nd harmonic3rd harmonic
Bipolar single stage$n_i/(4D)$$n_i^2/(12D)$
FET single stage$n_i/(8D)$Ideally zero
Bipolar differential stage$(m+c) \, n_i/(4D)$$n_i^2/(12D)$

Open and Closed Loop Frequency Response

The open-loop frequency response is the frequency response of the amplifier with no feedback — before feedback, or with the feedback network deliberately broken.

The closed-loop frequency response is the frequency response of the amplifier with feedback.

These two are closely related. The theoretical closed-loop frequency response is equal to the open-loop frequency response times the amount of feedback. If you have 40dB (100 times) of feedback, and an open-loop response of 1kHz, the closed-loop frequency response is 100kHz.

The formula relating these two is:

$$f_{cl} = \frac{f_{ol}}{1 + A_{ol} D}$$

where $D$ is the feedback factor and $A_{ol}$ is the open-loop gain. The total denominator expression is what we call feedback.

The open-loop frequency response is determined by the internal compensation (intended or not) of the amplifier. Many amplifiers are designed with one stage having a very high output impedance, so the stray capacitance of that stage’s output determines the open-loop frequency response. For integrated circuits the open-loop frequency response is either specified, or you can see it graphically as a function of gain — in the latter case, look at the maximum gain, which means zero feedback.

Just to remind you: the closed-loop gain $A_{cl}$ is related to the open-loop gain $A_{ol}$ in exactly the same way as the frequency response, although inversely.

$$A_{cl} = \frac{A_{ol}}{1 + A_{ol} D}$$

At Electrocompaniet the thinking favored a large open-loop bandwidth. This is also my opinion, but I feel it shouldn’t be larger than necessary. There is always a tradeoff, and if you go for too high an open-loop bandwidth, you reduce the possible amount of feedback you can have. My thinking is that as long as the open-loop bandwidth is high enough, you should use the rest of your gain for feedback. This will give you a more optimal design, because the overall distortion will be reduced.

What determines the open-loop bandwidth

Mostly it is determined by the last voltage-amplification stage. The collectors of this stage (assuming transistor amplifiers, which these articles are all about :-)) are connected to the bases of the drivers of the output stage. The input impedance of these drivers is normally very nonlinear, and strongly frequency-dependent. This means you can very well get a major pole here which varies strongly with signal level and the load (loudspeaker and cables) of the output stage. The solution to this is to voltage-drive the output stage, thus loading down the amplification stage. This will also have the effect of pushing up the cutoff frequency at this point. The EC amplifiers have this pole around 500kHz. The benefits of a voltage-driven output stage are described elsewhere on this site, including in an AES paper (not yet ported to this site).