I was once struck by one phrase from the book of Bob Cordell , about the same content: "Lovers of tube sound prefer the sound of lateral field-effect transistors". There was a question – how can you distinguish the sound of an amplifier with lateral transistors, if dazzles in the eyes from the number of zeros after the decimal point in the distortion of both amplifiers?
The problem is, that in the technical characteristics of the UMZCH distortions are indicated not for a real signal under a real load, and for a pure sinusoid with a frequency of 1 kHz into resistive load. In this case, measurements are made on a signal of large amplitude, masking almost all major problems, arising near zero. Everything turns out to be great for lovers of rumbling "heavy metal". And they, who wants to hear a woman whispering against the background of a double bass or the subtlest intonations of a voice, are forced to buy tube amplifiers with a maximum power 10 W at a cost of one thousand USD per watt or heat the room with transistor amplifiers, working in class A.
All problematic questions, concerning the coordination of the positions of the manufacturers of UMZCH, which lure consumers with extremely low Kg (менее 0.0001%, but at the frequency 1 kHz) and music lovers, boil down to three problematic positions, which can be denoted as: zero problem (switching distortion), load problem, and test problem.
It occurs in push-pull power amplifiers near zero load current, when there is a transfer of control of the output current from one arm of the amplifier to another. If the output transistors have a certain quiescent current, then when the signal crosses zero, a through current appears, passing through both transistors, the amplifier goes into class A for a short time, its internal resistance falls, transmission ratio is increasing.
As the output current rises, one of the output transistors gradually turns off, and the second, still at the start of the input characteristic, begins to fully determine the output current and, if the quiescent current is relatively small, his internal resistance (inverse slope at a given current) falls sharply. Consequently, for normal operation in this mode, the quiescent current of the transistors should be quite significant (class AB).
For this reason, most monographs authors [1-4] recommends the mode of operation of the output stage in a mode close to class B, when each arm of the amplifier works out exactly half of the signal, and the other half – is in the "cutoff". This is achieved by selecting the optimal смещения on the bases of the output transistors, when the signal is no longer "cut", but there is still no noticeable through current.
By default it is assumed, that when the output voltages of the amplifier are close to zero, the relative amount of distortion (kg) can be as big as you like, since the absolute value of distortion is obtained by multiplying them by a small value of the output signal and the result is negligible. Everything turns out fine with a large signal and most importantly, with resistive load. The problem is, that when the signal grows near zero, when the output transistors have just exited the current cutoff state, amplifier with deep feedback is forced to go to the maximum operating mode, forcing the opening of the transistor. At this moment, the gain drops sharply, the output impedance increases, transient distortion occurs, which are fundamentally impossible to eliminate using feedback.
This is demonstrated by simple, rough estimates. Really, slew rate of the output signal with an amplitude of about 30 And frequency 20 kHz reaches 4 V / ms. Amplifier output current 100 mA at load 4 Ohm is reached at the output voltage 0.4 In time 0.1 microsecond. [0.4 AT/(4 V / ms)= 0.1 μs]. During this time, the preamplifier must have time to overclock the output transistor, brake in time and provide full current control mode.
It means, that the settling time of the transient response of the entire amplifier in the small-signal mode should be, anyway, no worse than about 0.1 microsecond, which is an order of magnitude less, what do we have in real designs of UMZCH. In this way, we get a "dead zone", in which the properties of the output signal no longer depend on the input signal, but from the internal properties of the amplifier. In fact, the amplifier for a short time goes into a pulse mode of operation., when its properties are determined by transient characteristics, depending on the amplitude-phase-frequency characteristics in the high-frequency region (in megahertz range). Distortion signal spectrum, arising in such processes, also extends to the megahertz region.
Further we face the second problem, which is often ignored, – load problem. Some (For example, BUT. Danilov ) утверждают, that the resistive load is driving the amplifier, anyway, no less than real, other (For example, Sloan ) cite lack of load standards, as if it's an argument, to take advantage of the load, not corresponding to the real. certainly, operation on a resistive load in a certain respect reflects the quality of the amplifier, but an understanding of the general problem is still necessary. The reality is this, that the load in the form of a dynamic loudspeaker head is a fundamentally different type of load compared to a resistive one.
The main difference (besides inductance, which is less significant) consists in the presence of back-EMF head, proportional to voice coil speed (ZK) and directed against the emf of the amplifier. Consider the option, when the signal frequency is above the frequency range of the resonant peak of the head.
In this case, during the action of one of the signal half-waves, the ZK accelerates, and the maximum speed value is reached exactly at the moment of zero of the output signal – the voice coil speed signal lags behind 90 degrees from acceleration signal. At the same moment (zero output voltage) the back-EMF current reaches its maximum value, and this current must be extinguished by the internal resistance of the amplifier. note, that the back-EMF current is very significant and may even exceed the operating output current of the amplifier. Besides, with a real signal, the phase shift between the amplifier current and its output voltage leads to the fact that, that the zero crossing point of the output current UMZCH "walks" in the full range of output voltages, not at zero, as with resistive load. The problem of zero becomes the problem of the whole UMZCH.
A similar mismatch between the nature of the load and the type of input signal creates a test problem. Test signal, most relevant to the real, should be white noise, or a noise signal, DIN standardized. Нагрузка – this is a dynamic head, and distortions are determined by the method of compensating the output signal UMZCH with an input signal in the full operating range of output signal levels. This gives the total vector error, taking into account nonlinear, intermodulation and phase distortion.
If the magnitude of the distortions obtained during such a test falls within 0.1 %, then we can say, that the error, amplifier introduced, is out of range at least, middle ear. Such tests will help to reveal the true value of claims for "impeccability" or even "absoluteness" UMZCH.
The root of the problems listed – in the low quality of the output stage UMZCH. Cascade, crossover-free, having low distortion of all kinds in the audio frequency range, spectrum of harmonics insignificant in width, freely working on reactive load of any kind – such a cascade removes these problems. It is about such a cascade that will be discussed below..
1. Situation analysis and problem statement.
It should be emphasized: the presence of a "dead zone" in the behavior of the output transistors – the problem is not these transistors, but the whole scheme, working with OOS, since at these moments the entire amplifier forces the amplification to the limit, and then for a long time "comes out of shock". Voltage surges, limiting amplification, and rattlesnake, accompanying exit from this mode, will be observed at any points of the UMZCH. Installation of output inductors to suppress surges going to the load – this is burying your head in the sand, since outliers do not disappear in the circuit itself. An emitter or source follower never generates spikes or bounce by itself, because it works in full accordance with its capabilities without afterburner and overloads.
The first criterion for the normal operation of the amplifier – no voltage surges or bounce, indicating an overload of individual circuit elements in the operating range of frequencies and amplitudes. These overshoots should be absent in a pure distortion signal., since in the complete signal they are masked by the main amplitude.
Second quality criterion – distortion factor should not increase with decreasing output signal amplitude. Much better, if it decreases.
Third criterion – the amplifier should work normally on a reactive load (capacitance or inductance) creating at some frequencies a resistance equivalent to ohmic.
Douglas Self, researched in detail the transient distortion in the output stages , followed the path of finding the optimal bias on the bases of the output transistors, at which the distortion of the output stage will be minimal. In this case, the optimum found was related to the static (or quasi-static, that is, low-frequency) cascade operating mode, for high amplitudes and resistive loads. He also showed, that with decreasing signal amplitude, relative distortions increase, however, their absolute value falls, which seemed to him quite acceptable.
However, on most charts, cited by the author, there is an increase in distortion in direct proportion to the frequency, beginning with 1 – 2 kHz. This indicates the presence of a dynamic component of distortion, whose role grows with frequency, since it cannot be effectively suppressed by feedback. Therefore, the author's general conclusion about the advantages of class B, bias-optimized, should be attributed only to the low-frequency range, about 2 kHz. At higher frequencies, amplifiers take precedence., operating at significant quiescent currents. The dynamic component of distortion is associated with the transition of the preamplifier to the limiting modes, due to its inability to quickly transfer the output transistors to active mode.
Hun-Chan Lin is worthy of a monument – the structure of its UMZCH, published in the distant 1956 year, continues to dominate 21 century. but, improving cascades of preliminary amplification with the subsequent application of OOS is tantamount to using poultices to revive a dead body – feedback does not work on a closed or overloaded transistor.
Lin's root structure flaw – combination of control of the output stage for direct and alternating current. A signal of a certain polarity turns on the corresponding arm of the output stage and simultaneously turns off the other arm, which leads to the above phenomena.
Attempts to solve the problem within the framework of the same structure were reduced to the following simple principle – it is necessary to implement such a mode of operation of the output transistors, at which no current cutoff condition is allowed. Such publications appeared abroad in the early 1980s and this mode was later called Super A or Non switching. As I know, one of the first structures of this type in the Soviet space was proposed by Yu. Mitrofanov in 1986 year . He designated this mode as "economy mode A". The principle of organizing such an output stage is most clearly expressed in the amplifier G. Bragin . You should consider it in more detail., to identify the disadvantages of this approach.
Essential details, explanatory schemes, shown in Fig. 1. As, how the positive half-wave ends, current through resistance Ro falls, transistor T3 is opened with voltage across the diode, and the emerging current from the collector T3 is used (using the corresponding scheme) for a little bit- transistor T1 in order to maintain a certain level of output current for the entire period of the negative half-wave. When the positive half-wave comes again, transistor T1 is already half-open and is gently switched on without surges and overloads. measurement, conducted by the author, showed a significant narrowing of the spectral composition of harmonics.
It would seem that, everything turns out beautifully, except one – output current maintenance circuits are non-linear and operate autonomously, each in its own shoulder and only during the period of activity of the opposite shoulder. It backfires. According to Kirchhoff's rule, the sum of the currents in the node (at the amplifier output) It is always zero: 1+ – 1– – Iout = 0. Since the negative shoulder "knows nothing" about the processes in the positive shoulder, its current is proportional to the input signal KUin. We get Iout = 1+ – Than. The excess current of the positive shoulder has nowhere to go, except to go to the load, and it becomes a pure source of nonlinearity (distortion type meander). The newly created difficulty is heroically overcome due to the action of the general feedback.
We come to the conclusion, what's the extra current, passing through the output transistors must be the same for both arms. This must be a through current and the amplifier must maintain the same (not necessarily permanent)through current.
This mode is automatically implemented when using the aforementioned lateral field-effect transistors. [2SK1056(57, 58)/2SJ160(161,162), 2SK133(134,135)/2SJ48(49,50), BUZ901(900)/BUZ906(905)], switched on according to the scheme with a common source. The fact, that these transistors are characterized by the presence of a drain current at zero gate-source voltage (about 100 mA for pair 2SK1058 / 2SJ162). There is no current cut-off, and the "softness" of the input characteristics of these transistors – gradually increasing steepness and thermostable point near 150 mA allow, with a competent approach, to achieve good results by simple means. How visible, the choice of audiophiles is quite consistent with objective factors.
The situation is quite different with field-effect transistors with a vertical channel., in which the working area begins with a few volts of gate-source voltage. The trouble is compounded by the fact, that this type of transistor is technological, has a number of technical advantages (in particular, great steepness), wide range and low cost, because of what such transistors are massively installed in the output stages of the UMZCH.
BUT. Smooth popularizes amplifier circuits with separate amplification of half-waves and their subsequent addition at the output . In this version, good stabilization of quiescent currents can be obtained. (each arm acts as a current amplifier) and a fairly complete compensation of these currents when they are summed in the load. There are only two questions that arise – amplifier operation at low voltages (due to the splitting of the total signal into two half-waves using diodes), and – with reactive amplifier load (since each arm works with OOS for the output current, and the general OOS – by output voltage).
The nature of distortions in a push-pull UMZCH consists in a constant change in the operating mode of a powerful transistor – in one half-wave of the signal, its operating point goes through a cycle from cutoff to maximum current and back to cutoff. The problem is solved in class A amplifiers, where the operating point of the transistor remains virtually unchanged, and signal distortions are insignificant and are associated only with a changing voltage drop across the transistor (Earley effect). Mode AB gives some effect, provided, that the quiescent current of the transistor is significant (about 20% максимального). But still – this is a half measure: distortion increases, as soon as the opposite side transistor closes, and all the current of the operating transistor begins to flow into the load.
Research shows, that it is possible to ensure good and continuous controllability of the operation of the transistors of the output stage, if current passes through the output transistors always, and always exceeds the current, loaded. In this case, the harmonic composition of the distortion spectrum is gradually reduced to one – second harmonic.
In light of the above, we can formulate the requirements for the operating mode of a full-fledged output stage.
- currents, passing through the output transistors must exceed currents, passing through these transistors to the load, and it is desirable to provide the ability to control this excess current.
- The excess current must be the same (modulo) for both arms of the output stage even before switching on the feedback. (OOS equalizes these currents in any case, but this equalization is never complete). In other words, the circuit must provide a continuous and controlled through current through the output transistors and both transistors must be continuously in active mode. It should also take into account the presence of the dynamic component of distortion in the circuit design., depending on the slew rate of the output voltage.
2. circuit design.
First, you need to "cut the Gordian knot" of Lin's scheme – combination of control of the output stage for direct and alternating current. The signal amplification circuits and the operating mode control circuits must be separated and not interfere.. In this case, the output stage turns into a functionally separate unit (quadripole), well-defined, knowledge of which is necessary to build a general scheme.
The requirement of independence of the mode control severely limited the choice of circuit solutions, Moreover, provided only one single option – differential amplifier stage. Its peculiarity is, that the current signal, supplied to the emitters of the differential pair, is perceived by the amplifier as common mode interference and is suppressed with an efficiency of the order of 60 dB. In the same time, this signal controls the common-mode voltage across the collector resistors, and this voltage can be used to control the mode of powerful output transistors, connected according to the scheme with a common emitter (common east). Control is realized by using current sensors – resistors, installed in the load current circuit and, if necessary, then in the circuits of the output transistors. Actually, this is the idea of a circuit solution, which can be implemented in an unlimited number of options, depending on the specific task and the available element base. One of the possible options for constructing a circuit is discussed below..
The scheme was built as "short", to take full advantage of the frequency properties of the output transistors and avoid the appearance of additional poles of the frequency response. Besides, the differential stage enables a distortion reduction method using a compensating input current . This method does not use feedback and therefore does not create new frequency response poles..
The basic circuit of the output stage is shown in Fig. 2.. it – standard balanced differential amplifier circuit with high power output transistors. Its feature – in symmetrical quiescent current control (through flow) differential pairs of upper and lower shoulder – output Us (an increase in the quiescent current corresponds to a negative voltage at the input Vc). Actually, this control allows you to solve problematic issues of the output stages. The only question is the competent organization of this department.. Sensors, allowing to control the currents of the output transistors and the load current, are the resistors 0.1 ohm. The corresponding control points are marked with the letters A, B and C.
Initially, I focused on cheap high-power vertical channel field-effect transistors, which I had – pair IRF540 / IRF9540. Before picking up a soldering iron, I researched various circuit options in Micro-Cap10 environment.
Setting up the basic schema
The circuit in the variant in Fig. 2 is suitable for driving transistors., very steep (bipolar and field-effect transistors with vertical channel). If the slope is small, such as, in the case of high quality and expensive lateral field effect transistors, you can use a dynamic load or an additional amplification stage. Consider the general case of setting up a circuit for arbitrary output transistors.
First you need to decide on the gain and type of amplifier (inverting or non-inverting). Figure 3 shows the options for turning on the inverting (rys.Z,(a)) and non-inverting (Figure 3 b) amplifier for gain K = 5 with resistor values and frequency correction elements (to exclude self-excitation of the circuit).
Setting up the amplifier is reduced to determining the value of the resistors R0 (Fig. 2), determining the operating point of the output transistors with a grounded control input, Vс = 0. The quiescent current of the differential stage is set by the resistor R5 (R4) and is chosen quite significant, about 10 mA or approximately 5 mA on the shoulder. This current determines the required bias value of the output transistor., as the voltage drop across the resistor R0 minus the voltage drop across the junction of the transistor T9 (T10).
Finally, the quiescent current of the output transistors is determined by the value of the control voltage at the input Vc. Precise balancing of the amplifier is possible by changing the resistances R5 and R4 in this way, so that the output voltage becomes close to zero, which however, doesn't make much sense. The resistance of the resistors R2 determines the maximum closing speed of the output transistors, but its excessive decrease degrades the linearity of the amplifier and leads to excessive heat generation in T9 transistors, T10. For a pair of transistors IRF540 / IRF9540 with gain K = 5, the ratings of the circuit elements are as follows: R0 = 820 Ohm, R 1 = 10 ohm, R2 = 270 Ohms, R4 = R5 = 1.4 kΩ (at a quiescent current of about 70 mA), Scor = 68 pF, differential pair transistors, like the pre-final – medium power selected – T1, T2, T10 – BD139, T3, T4, T9 – BD140. The rest of the n-type transistors - 2N5551, and r-type - 2H5401. The type of transistors chosen was mainly determined by their availability at the developer, not specific requirements of the scheme.
It was produced in full accordance with the methodology, set out in article . Figure 4 shows the compensation circuits for both amplifier options.. Resistor ratings correspond to the condition, that the input current of the amplifier is negligible. Since this condition is not actually met, then the final correction of the gain of amplifier A2 was carried out by adjusting the feedback resistor Rfb + Rin to minimize signal distortion. Its optimal value turned out to be approximately 10% higher than the calculated one and amounted to 1650 Ohm instead 1500 ohm.
Figure 5 shows the graphs of the distortion signals up to (from above) and after connecting the distortion compensation circuit. The graphs show the output signal for two frequencies – 2 kHz and 20 kHz at a quiescent current of the output transistors of about 200 mA, supply voltage ± 30 V and output signal ± 22 V (on the charts he, Compare, reduced in 1000 time). Striking complete identity of the distortion signal for the basic amplifier, with so much (in order) differing frequencies. No wonder, since the amplifier has a large margin of bandwidth and audio range for it – this, in fact, quasi-static operation. Another matter, correcting amplifier A2, which I used NE5534. Since the spectrum of the distortion signal changes in direct proportion to the frequency of the main signal, then the ability to compensate for distortion decreases with increasing frequency. at a frequency 2 kHz distortion is suppressed in 68 time, and on 20 kHz – only in 23 fold. Naturally, the high-frequency part of the distortion is suppressed less, therefore, sharp spikes appear in the distortion signal.
The resulting amplifier by the type of frequency response approaches a single-pole with a pole of the 1st order. Pole frequency f(-45°)= 2.3 MHz, f(-180°)= 25 MHz f(0 dB)= 13.6 MHz, phase shift 0.26 ° by 20 kHz. The amplifier is resistant to loads of various types, in particular, capacitive in 2 uF, current equivalent resistor 4 Ohms at 20 kHz.
(To be continued)
Author: Pavel Cherednik, Volyn region. Shatsky district, with. Mills